A Flood Forecasting Method Based on Machine Learning Applicable to Watersheds with Lack of Runoff Data

Abstract

The invention discloses a machine learning-based flood forecasting method suitable for watersheds lacking runoff data, comprising the following steps: 1) sample watershed feature extraction and parameterization; 2) watershed flood response characteristics analysis; 3) generating watershed characteristic sample sets ; 4) Generate classification tree based on watershed feature sample set; 5) Generate training data set based on tree nodes; 6) Flood forecast based on classification tree and data-driven model; 7) Update classification tree and training set. Using machine learning algorithms to analyze the flood response characteristics of watersheds, based on the characteristics of watersheds and flood response characteristics, the association between watersheds is established. The invention generates sample data sets based on watershed characteristics and flood response similarities, and then trains the data-driven model based on the sample data sets. , to simulate the relationship between rainfall and flood response of small and medium rivers, so as to realize real-time flood forecasting of small and medium rivers. The method provided by the invention can realize the application of the data-driven model in the flood forecast of the watershed lacking runoff data, and change the dependence of the previous parameter transplantation on the model structure and model parameters, thereby improving the accuracy of the flood forecast.

Description

technical field [0001] The invention belongs to the technical field of water conservancy engineering, in particular to the technical field of flood control forecasting, and specifically relates to a machine learning-based flood forecasting method suitable for watersheds lacking runoff data. Background technique [0002] At present, my country's major rivers and their main tributaries have formed a flood control engineering system based on dikes, reservoirs, and flood storage and detention areas. Non-engineering measures such as flood control early warning and forecasting systems have also been gradually strengthened, which can basically prevent major rivers from flooding. However, for more than 50,000 small and medium-sized rivers, their distribution is wide, their number is large, their natural geography and climatic conditions are complex and diverse, and their flood control capabilities are generally backward. Especially in recent years, extreme weather events have increas...

AI Technical Summary

Problems solved by technology

However, data-driven models often require a large amount of rainfall and runoff data to train the model, so they have not been applied to flood forecasting in basins where runoff data is scarce.

Images